CODA

This repo contains codes for the following paper:

Jiaao Chen, Diyi Yang: Simple Conversational Data Augmentation for Semi-supervised Abstractive Conversation Summarization. EMNLP 2021

If you would like to refer to it, please cite the paper mentioned above.

Natural Language Undertanding

Prerequisite:

• CUDA, cudnn
• Python 3.7
• PyTorch 1.4.0

Pre-training the Utterance Generation Model

1. Install Huggingface Transformers according to the instructions here: https://github.com/huggingface/transformers.

2. Train the utterance generation model:

>>> chmod +x pre-training.sh
>>> ./pre-training.sh

Run CODA

1. Install Huggingface Transformers according to the instructions here: https://github.com/huggingface/transformers.

2.Train model with the Semi-CODA data augmentation strategies:

>>> chmod +x sf_train.sh
>>> ./sf_train.sh

Acknowledgement

This repository is built upon https://github.com/dinghanshen/Cutoff